{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XKNODIBJ2LJGWKS2SXX7KWJON2","short_pith_number":"pith:XKNODIBJ","schema_version":"1.0","canonical_sha256":"ba9ae1a029d2d26b2a5a95eff5592e6eaa900f0b39aa1670de70803f98d393dc","source":{"kind":"arxiv","id":"2607.23153","version":1},"attestation_state":"computed","paper":{"title":"In-Context Learning as Implicit Policy Gradient","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Masahiro Kaneko, Timothy Baldwin","submitted_at":"2026-07-25T11:13:52Z","abstract_excerpt":"Recent work has shown that large language models (LLMs) can iteratively improve their outputs by incorporating generated samples and their corresponding evaluation scores as in-context examples. Despite these empirical findings, the theoretical foundations underlying this phenomenon remain poorly understood. In this paper, we show that score-conditioned In-Context Learning (ICL) admits a structural correspondence to policy gradient optimization. We first provide a constructive proof that self-attention mechanisms can implement reward-weighted aggregation analogous to the REINFORCE algorithm un"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2607.23153","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-25T11:13:52Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"9ba589bec8ff077c48f70dff96808e237179d0d48f04fe2fb83a03ac7660e004","abstract_canon_sha256":"d91142588680578a815bf590dc4c19690a6d4ca55e4bd41457a9abcde14f1620"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T01:22:09.662687Z","signature_b64":"bQHCxFSTi6kWd9LDEKKsqzmw4Ay3XTCIZBpHUgW0dNwaD5Rigw2I6fxBZJGGBMU/3N2FF5skhw56cHOWvjK/DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ba9ae1a029d2d26b2a5a95eff5592e6eaa900f0b39aa1670de70803f98d393dc","last_reissued_at":"2026-07-28T01:22:09.661809Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T01:22:09.661809Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"In-Context Learning as Implicit Policy Gradient","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Masahiro Kaneko, Timothy Baldwin","submitted_at":"2026-07-25T11:13:52Z","abstract_excerpt":"Recent work has shown that large language models (LLMs) can iteratively improve their outputs by incorporating generated samples and their corresponding evaluation scores as in-context examples. Despite these empirical findings, the theoretical foundations underlying this phenomenon remain poorly understood. In this paper, we show that score-conditioned In-Context Learning (ICL) admits a structural correspondence to policy gradient optimization. We first provide a constructive proof that self-attention mechanisms can implement reward-weighted aggregation analogous to the REINFORCE algorithm un"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.23153","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2607.23153/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2607.23153","created_at":"2026-07-28T01:22:09.662250+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.23153v1","created_at":"2026-07-28T01:22:09.662250+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.23153","created_at":"2026-07-28T01:22:09.662250+00:00"},{"alias_kind":"pith_short_12","alias_value":"XKNODIBJ2LJG","created_at":"2026-07-28T01:22:09.662250+00:00"},{"alias_kind":"pith_short_16","alias_value":"XKNODIBJ2LJGWKS2","created_at":"2026-07-28T01:22:09.662250+00:00"},{"alias_kind":"pith_short_8","alias_value":"XKNODIBJ","created_at":"2026-07-28T01:22:09.662250+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2","json":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2.json","graph_json":"https://pith.science/api/pith-number/XKNODIBJ2LJGWKS2SXX7KWJON2/graph.json","events_json":"https://pith.science/api/pith-number/XKNODIBJ2LJGWKS2SXX7KWJON2/events.json","paper":"https://pith.science/paper/XKNODIBJ"},"agent_actions":{"view_html":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2","download_json":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2.json","view_paper":"https://pith.science/paper/XKNODIBJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.23153&json=true","fetch_graph":"https://pith.science/api/pith-number/XKNODIBJ2LJGWKS2SXX7KWJON2/graph.json","fetch_events":"https://pith.science/api/pith-number/XKNODIBJ2LJGWKS2SXX7KWJON2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2/action/storage_attestation","attest_author":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2/action/author_attestation","sign_citation":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2/action/citation_signature","submit_replication":"https://pith.science/pith/XKNODIBJ2LJGWKS2SXX7KWJON2/action/replication_record"}},"created_at":"2026-07-28T01:22:09.662250+00:00","updated_at":"2026-07-28T01:22:09.662250+00:00"}